Sugarcane harvest progress monitoring for commodity-supply timing
Multispectral time series track the week-by-week spatial advance of sugarcane harvesting across mill supply zones, converting NDVI collapse into raw-sugar supply forecasts before official statistics appear.
Sensors
- Sentinel-2 MSI: 10 m resolution in red, NIR and SWIR bands; 5-day revisit at the equator with both satellites combined. Red-edge bands (B5, B6, B7 at 20 m) sharpen canopy-closure detection and separate green ratoon regrowth from standing mature cane.
- Planet SuperDove: 3 m resolution, daily revisit over tasked areas. Eight spectral bands including two red-edge channels. Sufficient to resolve individual field parcels of 0.5 ha or smaller, reducing boundary-contamination error in mill-zone aggregations.
- Landsat 8/9 OLI: 30 m resolution, 16-day revisit per satellite (8-day combined). Free archive back to 1972 for Landsat programme as a whole. Useful for multi-year harvest-calendar baselines and inter-annual comparison of harvest pace.
- MODIS Terra/Aqua: 250 m in red and NIR bands (bands 1 and 2), daily revisit. Too coarse to resolve individual fields in fragmented landscapes but valuable for regional-scale harvest-progress curves over large, consolidated Brazilian cerrado plantations where field blocks exceed 50 ha.
What a harvested cane field looks like from orbit
Mature sugarcane carries a dense green canopy that drives NDVI values above 0.7 in the 60 to 90 days before harvest. Mechanical harvesting strips that canopy in a single pass, exposing bare red-brown soil or low stubble within hours. NDVI drops to values below 0.2, sometimes below 0.1 on freshly cut fields. That collapse is abrupt and spatially coherent: a harvested block changes state across its entire area within one or two satellite overpasses, not gradually.
The signal is unambiguous in multispectral imagery when cloud cover cooperates. Sentinel-2's red-edge bands add a useful check: they distinguish the pale spectral signature of cut stubble from the green flush of ratoon regrowth that begins within two to three weeks of harvest. Catching the harvest event rather than the subsequent regrowth requires either high revisit frequency or cloud-free conditions during the detection window, which is the central operational challenge in wet-season producing regions.
From field events to mill-zone supply curves
The analytic architecture has three layers. First, field boundaries are delineated, either from cadastral data, from automated parcel segmentation on high-resolution imagery, or from a combination of both. Second, each field is assigned to a mill catchment using published or client-supplied mill-zone polygons. Third, a time series of NDVI observations is maintained per field, and each field is classified as standing, harvested or regrowing based on threshold crossings and temporal shape rules.
Aggregating harvested area week by week across a mill's supply zone produces a harvest-progress curve. Multiplied by an estimated yield per hectare (drawn from historical remote-sensing models or ground-calibration data), this curve becomes a weekly estimate of cane available for crushing. The resulting supply forecast runs two to four weeks ahead of what mills report to national statistical agencies, which is the window that matters for raw-sugar futures positioning and physical procurement decisions.
Brazil's centre-south region, which accounts for roughly 90 percent of Brazilian production and a large share of global traded raw sugar, runs its harvest from April through November. Australia's Queensland mills operate May through December. India's Maharashtra and Uttar Pradesh belts harvest October through March. The seasonal windows are staggered enough that a global monitoring programme can sustain year-round analyst attention without duplicating effort.
Resolution, revisit and the cloud problem stated plainly
Sentinel-2 at 10 m resolves fields above roughly 0.5 ha without significant boundary contamination. Planet SuperDove at 3 m pushes that floor to around 0.1 ha, which matters in India's smallholder belts where field fragmentation is high. Landsat 9's 30 m pixels introduce mixed-pixel error on field edges but remain adequate for large consolidated plantations.
Revisit is the harder constraint. Sentinel-2's nominal 5-day revisit degrades in practice when cloud cover invalidates acquisitions. During the wet-season overlap in parts of India and Queensland, usable observations can drop to one every 10 to 20 days, compressing the detection window for individual harvest events. Ratoon regrowth restores NDVI above 0.4 within three to four weeks of harvest, so a missed observation can mean a harvested field is reclassified as standing by the time the next clear image arrives. MODIS's daily revisit partially compensates at the regional scale but cannot resolve individual fields. SAR sensors are not listed here because this page focuses on optical methods; cloud-penetrating SAR approaches are covered in the SAR-focused sibling pages.
Calibration anchors and honest uncertainty ranges
A harvest-progress curve is only as useful as its yield assumption. Remote sensing measures harvested area with reasonable confidence; it does not directly measure tonnes per hectare. Published studies using Sentinel-2 time series in Brazil's São Paulo state have reported harvested-area classification accuracies above 85 percent at the field level under good cloud conditions, but accuracy degrades in fragmented landscapes and during cloudy periods.
Yield per hectare varies with variety, ratoon cycle number, irrigation status and weather. A model that applies a single regional average yield will produce supply estimates with uncertainty bands of 10 to 20 percent or wider in heterogeneous regions. Narrowing that range requires either ground-truth yield data from mills or integration with crop-growth models calibrated to local conditions. Buyers of this analysis should treat the output as a directional leading indicator, not a certified inventory figure. The value is in tracking relative pace and spatial distribution of harvest progress, not in claiming tonne-level precision.
Where this fits in a commodity intelligence workflow
Raw-sugar supply forecasts derived from satellite harvest-progress data are most useful when combined with mill-crush-rate data (often available from industry associations with a lag), shipping and port data, and weather forecasts for the remaining harvest window. Satellite data fills the spatial gap: it shows which parts of a catchment have been harvested and which have not, something no survey or mill report provides at weekly cadence.
For commodity traders, the relevant output is a weekly percentage-complete figure per mill zone, updated as new clear-sky acquisitions arrive. For insurers writing crop revenue policies, the same curves provide an independent check on insured yields and harvest timing. For lenders with exposure to mill operators, harvest-progress data offers early warning of a slow or interrupted season before financial reporting reflects it.
Satellize runs multispectral crop-monitoring analytics on open constellations including Sentinel-2 and Landsat, with commercial tasking added on client licence. The methodology applied in the Kingdom of Tonga crop-estimation programme, which uses NDVI time series to track crop-area and condition, shares its analytical foundations with the mill-catchment approach described here.
Typical figures
| Spatial resolution (primary) | 10 m (Sentinel-2 MSI, red/NIR/SWIR bands) |
| Spatial resolution (high-resolution option) | 3 m (Planet SuperDove, 8 spectral bands) |
| Revisit cadence | 5 days nominal (Sentinel-2, both satellites); daily (Planet SuperDove tasked areas); 16 days per satellite / 8 days combined (Landsat 8+9) |
| Spectral bands used | Red (665 nm), NIR (842 nm), red-edge (705, 740, 783 nm), SWIR (1610, 2190 nm) on Sentinel-2 MSI |
| Minimum resolvable field size | ~0.5 ha at 10 m (Sentinel-2); ~0.1 ha at 3 m (SuperDove) |
| Harvest-event detection window | Approximately 3 weeks post-cut before ratoon regrowth restores NDVI above 0.4 |
| Update latency (operational) | 2 to 5 days after satellite acquisition, depending on cloud screening and processing pipeline |
| Archive depth | Sentinel-2 from 2015; Landsat from 1972 (30 m); MODIS from 2000 |
| Coverage | Global; operational focus on Brazil centre-south, Queensland (Australia), Maharashtra and Uttar Pradesh (India) |
| Delivery formats | GeoTIFF harvest-state rasters, mill-zone CSV progress tables, GeoJSON field-classification layers, PDF weekly briefings |
Analytics Satellize can run
| Weekly harvest-progress percentage per mill catchment | NDVI threshold-crossing detection in Sentinel-2 or SuperDove time series, aggregated over client-supplied mill-zone polygons | CSV or JSON feed updated on each clear-sky acquisition; cumulative-harvest-area chart per zone |
| Field-level harvest-state classification map | Per-parcel NDVI time-series classification (standing / harvested / regrowing) using temporal shape rules and red-edge band ratios | GeoJSON or GeoTIFF layer, refreshed weekly, with field-level state attributes and last-observation date |
| Estimated cane available for crushing | Harvested-area estimate multiplied by regional yield model; uncertainty bounds reported explicitly | Weekly PDF briefing with supply curve, confidence interval and narrative commentary on pace versus prior-year baseline |
| Harvest-pace anomaly alert | Comparison of current-season harvest-progress curve against 5-year historical average derived from Landsat and Sentinel-2 archive | Email or API alert when cumulative harvested area deviates more than one standard deviation from the historical mean for that calendar week |
| Cloud-gap interpolation and data-quality flag | Linear or harmonic interpolation across cloudy periods; per-field observation-count and cloud-fraction metadata attached to all outputs | Quality-flag layer accompanying each classification raster; gap-filled NDVI time-series CSV per field parcel |
| Multi-season harvest-calendar baseline | Retrospective NDVI time-series analysis over Landsat 8/9 and Sentinel-2 archive to characterise typical harvest start, peak and end dates per mill zone | Historical harvest-calendar report (PDF and GIS layers) covering up to 10 seasons, used to contextualise current-season pace |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.